Acknowledgement: the data for this Tidy Tuesday comes from Dan Oehm (https://github.com/doehm/alone)

How has viewership changed over the seasons

## `geom_smooth()` using formula 'y ~ x'

Number of viewers per episode of Survivor

Between seasons 1 and 6, viewership of Survivor generally trended downwards. Following an all-time low in viewership in season 6 (s6 ep 9 is the only episode in the series’ history to have fewer than 1 million US viewers!), the show has experienced a rebound to levels not seen since season 5 but it’s a long way to go before it scales the heights of the season 1 finale.
(The plot above is interactive!)

Interestingly, seasons 6 & 7, the two seasons with the lowest average number of viewers per episode, were both held at Great Slave Lake in Canada.
Average viewership per episode over the first 8 seasons of Survivor in the US

Average viewership per episode over the first 8 seasons of Survivor in the US

Viewership over the first 8 seasons of Survivor in the US

Viewership over the first 8 seasons of Survivor in the US



Why did viewership fall over the first 6 seasons

Perhaps the quality of the show declined?

No! The IMDB rating for Survivor over the first 6 seasons were more or less the same. In fact, the most highly rated season was season 6, despite it having such a low viewership. (This is assuming the IMDB rating is a reliable measure of quality!)
Average IMDB ratings for episodes in the first 6 seasons of Survivor

Average IMDB ratings for episodes in the first 6 seasons of Survivor



Perhaps the show became less exciting/dangerous?

Assuming the number of survivalists is a measure of how exciting a season was, viewership may have declined because the show became less exciting to watch as contestants found themselves in less perilous situations.

It does not seem to be the case that more survivalists being medically evacuated is correlated with a higher viewership. This makes sense as the IMDB ratings across seasons are very similar.

If anything, there seems to be a possibility that having fewer survivalists being medically evacuated is correlated with a higher viewership.
Number and percentage of medically evacuated survivalists and average viewership over 6 seasons of SurvivorNumber and percentage of medically evacuated survivalists and average viewership over 6 seasons of Survivor

Number and percentage of medically evacuated survivalists and average viewership over 6 seasons of Survivor



It does indeed appear that there could be some correlation between survivalists not being medically evacuated and viewership.
Number and percentage of survivalists NOT medically evacuated and average viewership over 6 seasons of SurvivorNumber and percentage of survivalists NOT medically evacuated and average viewership over 6 seasons of Survivor

Number and percentage of survivalists NOT medically evacuated and average viewership over 6 seasons of Survivor


Running Pearson’s correlation test indicates that the positive correlation between viewership and surivivalists not being medically evacuated is significant.

## 
##  Pearson's product-moment correlation
## 
## data:  notEvacuatedBySeason$mean_viewership and notEvacuatedBySeason$`Number of survivalists not medically evacuated`
## t = 2.4286, df = 6, p-value = 0.05126
## alternative hypothesis: true correlation is not equal to 0
## 95 percent confidence interval:
##  -0.001189892  0.941586313
## sample estimates:
##       cor 
## 0.7040734
## 
##  Pearson's product-moment correlation
## 
## data:  notEvacuatedBySeason$mean_viewership and notEvacuatedBySeason$`Percentage not evacuated`
## t = 3.8115, df = 6, p-value = 0.008848
## alternative hypothesis: true correlation is not equal to 0
## 95 percent confidence interval:
##  0.3354273 0.9705668
## sample estimates:
##       cor 
## 0.8412556

Could it be that viewers prefer the show when the health of participants is in less danger?



Predicting how many viewers season 10 might get

Number of viewers per episode of Survivor

Number of viewers per episode of Survivor

There appears to be a significant autocorrelation for 13 episodes suggesting that a high number of viewers for episode x will likely see a high number of viewers for episode x+13.
Autocorrelation of viewership

Autocorrelation of viewership


However, as demonstrated by the partial autocorrelation, there is likely noise associated with correlations from previous episodes driving up the autocorrelation coefficient at each lag. Before fitting a model, the data needs to be transformed so that the mean and variance are stationary.
Partial autocorrelation of viewership

Partial autocorrelation of viewership


Following transformation, we can apply an ARIMA model to the data and predict how many viewers the next season of Survivor will obtain!

Episode Predicted number of viewers in the US (millions)
1 1.437960
2 1.522137
3 1.474678
4 1.483932
5 1.527913
6 1.492780
7 1.493292
8 1.487127
9 1.479134
10 1.558419
11 1.513843